ArchAI

Understand the code. Check the rules. Prove the findings.

ArchAI checks a Python repository against engineering standards you configure, and cites the source evidence (file and line) for every finding. The repository knowledge, rules and findings can be exported as an Open Knowledge Format (OKF) v0.2 bundle and browsed or re-checked from that bundle alone.

Its rule is Receipts or Silence: a finding is a verified violation only when every supporting fact is cited from source. When static analysis cannot settle a question, ArchAI says so instead of guessing.

Install

git clone https://github.com/Hami0095/ai-architecture.git
cd ai-architecture
pip install -e .

Python 3.10+. A license token (or ARCHAI_DEV_MODE=1 for local development) is required; see License. Ollama is only needed for the LLM-assisted commands.

First commands

ai-architect CHECK-STANDARDS path/to/repo --okf-out bundle --preview
ai-architect OKF-VIEW bundle findings

Commands

	Command
Deterministic	`CHECK-STANDARDS`	Check standards with cited evidence; JSON, Markdown and OKF output
	`OKF-VIEW`	Browse any OKF bundle in the terminal
LLM-assisted	`AUDIT`, `IMPACT`, `G-REASON`, `EXPLAIN`	Exploration and explanation; not verified
Integrations	`GITHUB`	Repositories, pull requests, diff validation
Session	`CONFIG`, `SET-GITHUB-TOKEN`, `SET-PM-TOKEN`, `TRACE`, `HELP`

Sprint planning (PLAN, SIMULATE, RELEASE-CONFIDENCE) comes from ArchAI's sprint-planning case study; see Sprint-planning commands.

Documentation

CLI Command Reference: every command, argument and exit code.
Standards & Rules: configuration, the five rule kinds, finding statuses, JSON and OKF output.
Configuration & Environment: license, modes, config files, environment variables, LLM runtime.
Recipes: CI gating, investigating findings, sharing bundles, reproducibility.

What ArchAI does not do

Analyze languages other than Python (they are reported as not covered).
See runtime wiring such as dependency injection or plugins.
Measure test coverage.
Verify anything produced by the LLM-assisted commands.